Compose: docker-compose run --rm does not remove volumes

[ajc161]

Running one-off commands with docker-compose does not delete volumes used by the container. This is different than docker run --rm which does remove volumes after the container is deleted.

[yordis]

:+1:

[schmunk42]

We also ran into this.
What's very strange is that we had a cronjob running for months without any problems, but now every docker-compose run --rm leaves behind a volume with ~200 MB.

I am not 100% sure if this is related to an upgrade from 1.2.0 to 1.5.2, but it's the only change we've made.

docker info
Containers: 14
Images: 198
Storage Driver: aufs
 Root Dir: /var/lib/docker/aufs
 Backing Filesystem: extfs
 Dirs: 506
 Dirperm1 Supported: true
Execution Driver: native-0.2
Logging Driver: json-file
Kernel Version: 3.16.0-38-generic
Operating System: Ubuntu 14.10
CPUs: 2
Total Memory: 7.798 GiB
Name: ro109
ID: 4I66:JXGX:AAGY:ABRV:X7FD:ADPZ:IPHK:42BS:EUA2:QDRD:CJI3:EX6U
WARNING: No swap limit support

 Version:      1.8.3
 API version:  1.20
 Go version:   go1.4.2
 Git commit:   f4bf5c7
 Built:        Mon Oct 12 18:01:15 UTC 2015
 OS/Arch:      linux/amd64

Server:
 Version:      1.8.3
 API version:  1.20
 Go version:   go1.4.2
 Git commit:   f4bf5c7
 Built:        Mon Oct 12 18:01:15 UTC 2015
 OS/Arch:      linux/amd64

docker-compose version
docker-compose version 1.5.2, build 7240ff3
docker-py version: 1.5.0
CPython version: 2.7.9
OpenSSL version: OpenSSL 1.0.1e 11 Feb 2013

CC: @Quexer69

[dnephin]

That's strange, we've never set v=True for this rm, and I don't think that it's ever defaulted to false, so I'm not sure how the volumes were being removed before.

[schmunk42]

After digging through some code, I think it's very likely that someone ran also https://github.com/chadoe/docker-cleanup-volumes in a cronjob on the server.

Is there another way to remove the volumes from a docker-compose run?
Or would you recommend docker exec?

[dnephin]

It seems a bit strange that a one-off container would use a volume and just remove it immediately. I would think that it would easier use a host volume, or a named volume that stays around.

Is the volume just being created because the image has a volume in it?

docker exec would be one way around that. Implementing this feature would also solve it. I believe it's a small change to run_one_off_container() by passing v=True to project.client.remove_container().

[mixja]

This is a fundamental docker behaviour - if the image has volumes in it, they will created whenever you run a container from that image.

It's common to run one off tasks from an image that has volumes you may never use for that task.

The docker-compose run --rm behaviour should mirror the docker run --rm behaviour, which does correctly remove volumes.

[schmunk42]

@dnephin Our use-case is a rather simple web-app. We usually have web (nginx), php and worker (same image as php).
We're sharing static-files (assets) between web and php, but in other scenarios we also need to share files between php and worker.

The volume is defined in docker-compose.yml - sure there could be optimizations how much files are shared, but it still looks like a general problem to me.

docker exec could be a workaround, but I'd like to stay with docker-compose and I would rather not rely on the fact that my container is named project_php_1 because it may be named project_php_2 in some cases.

I also noticed that we ran a cleanup script before, I had to reactivate/fix that, but having an option to remove volumes after docker-compose run would still be great.

I need to look into named volumes a bit more I think, basically all apps we run, are running on a swarm, so I need to configure that properly.

[binary-data]

I also have this problem. I am using Docker Compose 1.7.1

[ebr]

same exact use case as @schmunk42. we use named volumes. docker-compose down does not remove the named volumes, which is preventing us from doing a complete clean up of artifacts on build. We are can workarounds like docker volume ls | grep myassets, but that is unmaintainable in many ways.

[jmenga]

You can work around this issue in Docker Compose 1.7 as follows:

docker-compose run xxx
docker-compose down -v

The key here is to not use the --rm flag on the run command. Because Docker Compose 1.7 removes containers started with the run command, it cleans up everything correctly including all volumes (as long as you use -v)

[jmenga]

@schmunk42

docker-compose down -v always removes named local volumes (1.6+), the current issue relates to volumes defined in the image of the service you are running that are automatically created not declared as explicit volumes in your docker-compose.yml.

For example given the following docker-compose.yml file:

version: '2'

volumes:
  mysql_run:
    driver: local

services:
  db:
    image: mysql:5.6
    volumes:
      - mysql_run:/var/run/mysqld

If we use docker-compose run --rm:

$ docker volume ls
DRIVER              VOLUME NAME
$ docker-compose run --rm db
Creating volume "tmp_mysql_run" with local driver
error: database is uninitialized and password option is not specified
  You need to specify one of MYSQL_ROOT_PASSWORD, MYSQL_ALLOW_EMPTY_PASSWORD and MYSQL_RANDOM_ROOT_PASSWORD
$ docker-compose down -v
Removing network tmp_default
Removing volume tmp_mysql_run
$ docker volume ls
DRIVER              VOLUME NAME
local               a79c78267ed6907afb3e6fc5d4877c160b3723551f499a3da15b13b685523c69

Notice that the volume tmp_mysql_run is created and destroyed correctly, but we get an orphaned volume which is the /var/lib/mysql volume in the mysql image.

If we use docker-compose run instead:

$ docker volume ls
DRIVER              VOLUME NAME
$ docker-compose run db
Creating volume "tmp_mysql_run" with local driver
error: database is uninitialized and password option is not specified
  You need to specify one of MYSQL_ROOT_PASSWORD, MYSQL_ALLOW_EMPTY_PASSWORD and MYSQL_RANDOM_ROOT_PASSWORD
$ docker-compose down -v
Removing tmp_db_run_1 ... done
Removing network tmp_default
Removing volume tmp_mysql_run
$ docker volume ls
DRIVER              VOLUME NAME

Everything is cleaned up correctly...

[reiven]

Can this be closed now?

[nkovacs]

Is it fixed? My PR is still open.

[hholst80]

I also don't understand this issue. Using docker-compose down is not an option if we are running a service stack in the same name space as the "one off" command we want to clean up?

[nkovacs]

docker-compose down -v would bring down the service stack and delete named volumes. I want neither to happen.

[ryneeverett]

Pretty sure this issue was about run --rm deleting unnamed volumes, which indeed now seems to be fixed.

[nkovacs]

It's not fixed in 1.10, I just tried it, and it doesn't look like it's fixed in master either (v=true is still missing): https://github.com/docker/compose/blob/master/compose/cli/main.py#L985

The issue is that --rm does not delete unnamed volumes. It should.